Distributions of read and write noise margins in large CMOS SRAM arrays are investigated by directly measuring the bit-line current during bitline / wordline (write) or cell supply (read) voltage sweep in a 768Kb 45nm CMOS SRAM test-chip. Good correlation between write/read margin estimates through the bit-line measurements and the DC read SNM (RSNM) and I W measurements in small on-chip SRAM macros with wired-out storage nodes are demonstrated. Four common writeability metrics are correlated and compared. Array-level characterization of SRAM cell read stability and writeability allow fast and accurate characterization of highdensity SRAM arrays is scalable for capturing up to 6 standard deviations of parameter variations. Introduction Continued increase in the process variability is perceived to be a major challenge in future technology scaling. To satisfy the functionality of hundreds of millions of SRAM cells in current ondie cache memories, the design has to provide more than 6 standard deviations of margin to parameter variations. Margins have been estimated through TCAD simulations or by measuring DC read/write margins in small SRAM macros with wired-out storage nodes [1-2] and direct cell current measurement in large SRAM arrays [3]. In this work, SRAM cell read stability and writeability margins are estimated through bitline currents and compared to direct DC RSNM and I W measurements.Measurement Methodology Writeability: During a write operation, a low-going BL R voltage pulls down on the '1' storage node as the WL is held high until the trip point of the inverter formed by PR and NR is reached, and a bit cell is successfully written (Fig. 1a). In this work, the measurements of four commonly used writeability metrics are compared and correlated. The BL write margin (BLWM) is the maximum BL voltage able to flip the cell state and can be measured through monitoring the BL L current at the '0' storage (I MEAS1 in Fig. 1a) while ramping down the BL R voltage with WL held high. The voltage at BL R that induces a sudden drop in I MEAS1 is the write margin of the cell (Fig. 1b). Similarly, a WL write margin (WLWM) can be defined as the maximum of (V DD -V WL ) in a WL voltage sweep (Fig. 1c) while monitoring the change in the BL current at the '1' storage node (I MEAS2 in Fig. 1a). Write margin can also be measured as the writeability current (I W , as defined in Fig. 4c and [4]), or the write noise margin derived from the butterfly curves (BFC) [1].Read Stability: The read stability of an SRAM cell is conventionally quantified by the cell read static noise margin (RSNM) [5] and is highly sensitive to the cell supply. To characterize the read stability of an SRAM cell in a large array, BL current at the '0' storage node (I MEAS1 ) is monitored while ramping down the cell supply with the bit-lines pre-charged and WL driven by the nominal supply. The difference between the nominal supply and the cell supply causing I MEAS1 to drop gives the read margin (RM) of the SRAM cell under test (Fig. 1d).Array Imp...